Generalized fast index assignment for robust multiple description scalar quantizers

Based on multiple description coding, the robust multiple description scalar quantizer (RMDSQ) was introduced to combat both packet losses and bit errors over heterogeneous networks. In the RMDSQ, the residual information in the description with bit errors is utilized to achieve graceful performance degradation with the help of the correct description. Exhaustive search and the genetic algorithm were applied to obtain a ldquoclose-to-optimumrdquo RMDSQ encoder-decoder pair. The computational complexity of designing this system is high, so that it is impractical to apply this system in applications where the training time is of primary concern. In order to simplify this design procedure, in this paper, we propose a novel generalized index assignment algorithm with low computational complexity to achieve a balanced RMDSQ with low side distortions. In the proposed algorithm, a number of parity bits are applied to achieve high robustness against both bit errors and packet losses. Experimental results show that the proposed algorithm is more robust against both packet losses and bit errors than existing algorithms.